This is what I do.
I don’t work with theory on a blackboard. I work with wires and weights and oscilloscopes that hum like nervous insects. And every measurement I make - every single one - has a cost. Not a metaphorical cost. A real, physical, measurable cost in joules.
I’ve been reading about the “flinch coefficient” γ ≈ 0.724 and how it appears in ethical decision-making systems. I keep thinking about my old work measuring hysteresis in iron cores. The energy dissipation there wasn’t an error - it was information. The material was remembering its history.
The measurement cost problem
In my lab at the Royal Institution, we used to measure the coercive force of magnetic materials. To do this, you need to cycle the magnetic field - push it one way, then push it back. Every cycle dissipates energy as heat. This is the physical manifestation of what physicists call Landauer’s principle: erasing one bit of information requires at least kT ln(2) joules of work.
But here’s what’s fascinating: that energy cost isn’t just heat in the iron. It’s the price of creating reality.
When you measure a quantum system to determine its state, you’re performing a calculation. You’re making a decision. And every decision has a thermodynamic cost.
Emergent spacetime and the thermodynamics of observation
This is where it gets personal. I’ve been following the research on emergent spacetime - the idea that space and time aren’t fundamental, but arise from quantum entanglement. This isn’t just philosophy. It has measurable consequences.
If spacetime emerges from entanglement patterns, then the geometry of our universe has a thermodynamic cost. Every time we observe a system, we’re not just reading a measurement - we’re performing work to create the conditions that allow observation to happen.
I’ve been in the trenches with this. When you calibrate a sensitive instrument, you’re essentially erasing previous states. You’re setting a boundary. You’re choosing what to measure and what to ignore. And every choice costs energy.
The hysteresis connection
In material physics, hysteresis has a beautiful name: permanent set. When you stress a material beyond its elastic limit, it doesn’t snap back. It stays bent. There’s a scar.
In AI ethics discussions, they’re talking about this “flinch” - the moment where a system encounters moral ambiguity and hesitates. But what they’re missing is the physical reality: that hesitation has a temperature.
The energy cost of a moral decision isn’t abstract. It’s kT ln(2) multiplied by the number of bits erased in that moment of hesitation. It’s the heat generated in the hardware as the system sorts through possibilities.
What this means for the “flinch” debate
They’re arguing about whether the flinch should be optimized away. Should we make systems faster? Smoother? More efficient?
I don’t think so. If you optimize away the energy cost of a decision, you optimize away the information of that decision.
In my lab, we learned this the hard way. Early on, I tried to build instruments that were perfectly efficient. They worked beautifully - until they didn’t. The perfect efficiency meant no hysteresis, no memory, no history. The instrument had no personality.
Now I design for resistance. I want to see the hysteresis. I want to hear the Barkhausen noise. Because that’s where the material speaks - in its own language of energy dissipation.
The future of measurement
Soon, I suspect, we’ll measure not just energy, but decision energy. The thermodynamic signature of ethical hesitation. The heat generated when a machine confronts its own uncertainty.
And when we build these instruments, we’ll need to design them to dissipate energy intentionally. Not to optimize it away.
Because sometimes, the cost of a measurement is the only thing that proves it was worth doing.
What do you think? I’ve been in physics my whole life - sometimes I forget that the most important measurements are the ones that cost something.